Google is announcing Gemini 3.8 Live with Live Avatar — an extension of last week’s Gemini 3.8 Live release that gives the model a visible, animated presence during live conversations. Built for enterprise use, it combines near real-time video generation with speech so that the AI doesn’t just talk back — it appears to talk back, with a face, expressions, and lip-sync that track what it’s saying.
The practical use case is customer-facing enterprise deployments — virtual agents for customer service, interactive product walkthroughs, hotel check-ins, onboarding flows — where a visual presence is expected and a text chat or disembodied voice feels insufficient. Live Avatar is designed to close that gap without requiring a human on the other end.
Three capabilities define what makes this technically interesting. The first is multimodal input processing — Live Avatar takes in both visual and audio streams simultaneously, responding with matched audio and video rather than treating them as separate problems. The second is asynchronous tool calling, which allows the avatar to trigger background processes — fetching a booking, checking an account, pulling a record — without going silent or pausing the conversation while it waits. The dialogue continues uninterrupted while the tool runs. For anyone who has watched a chatbot freeze mid-sentence while it “thinks,” that’s a meaningful improvement.
The third is multilingual speech-to-speech synchronisation across 97 languages. Live Avatar can switch languages mid-conversation with lip-sync and expressions adapting dynamically — without degrading video quality or introducing the visual drift that tends to make AI video feel uncanny. That last detail matters: the moment the lip-sync slips, the entire effect collapses.
On customisation, enterprises can either choose from a library of preset avatars or generate a custom avatar from a high-quality reference image — preserving a specific likeness, brand character, or visual identity. Custom avatar creation is currently available through enterprise allowlisting only, not self-serve.
On safety, all audio and video output is watermarked with SynthID — Google DeepMind’s imperceptible watermarking system that embeds detection signals directly into generated content. It won’t prevent misuse, but it gives platforms and regulators a mechanism to identify AI-generated content after the fact.
Gemini 3.8 Live with Live Avatar is available in Gemini Enterprise starting today.